From 6ac5f83bb23e46518c3aaea33b4d3c3995d41abe Mon Sep 17 00:00:00 2001 From: alehel Date: Sun, 15 Nov 2026 00:44:35 +0100 Subject: [PATCH] feat(media,video,audio): add Sierra VMD demuxer and decoders Sierra VMD (Video and Music Data) is the video format of Sierra and Coktel Vision games of the 1990s. Written clean-room from the public format description at https://wiki.multimedia.cx/index.php/VMD, with ffmpeg as a black-box oracle. - craft-media: `vmd` feature, a demuxer for the 816-byte header and the table of contents; `Container::Vmd`, `CodecId::VmdVideo` and `CodecId::VmdAudio`. - craft-video: `vmd` feature, a decoder for 8-bit paletted pictures (rendering methods 1 to 3, LZ packing, palette changes). - craft-audio: `vmd-decode` feature, a decoder for 8-bit PCM and 16-bit DPCM sound, mono and stereo. - craft-media-testkit: a VMD fixture writer, since ffmpeg reads VMD but cannot write it. Tests compare the decoders with ffmpeg on generated files, and on real files when CRAFT_MEDIA_VMD_SAMPLES names a directory of them. 16-bit, 24-bit and Indeo 3 pictures, IMA ADPCM sound and the later header revisions are not decoded. --- README.md | 2 +- crates/audio/Cargo.toml | 8 + crates/audio/README.md | 1 + crates/audio/src/lib.rs | 15 +- crates/audio/src/vmd.rs | 178 +++++++++++++ crates/audio/tests/vmd.rs | 265 ++++++++++++++++++++ crates/media-testkit/src/lib.rs | 2 + crates/media-testkit/src/vmd.rs | 343 +++++++++++++++++++++++++ crates/media/Cargo.toml | 6 + crates/media/README.md | 56 ++++- crates/media/src/id.rs | 12 + crates/media/src/lib.rs | 5 +- crates/media/src/vmd/mod.rs | 405 ++++++++++++++++++++++++++++++ crates/media/src/vmd/tests.rs | 161 ++++++++++++ crates/media/tests/oracle_vmd.rs | 68 +++++ crates/video/Cargo.toml | 12 +- crates/video/README.md | 1 + crates/video/src/lib.rs | 4 + crates/video/src/vmd/mod.rs | 382 ++++++++++++++++++++++++++++ crates/video/tests/vmd/main.rs | 415 +++++++++++++++++++++++++++++++ 20 files changed, 2329 insertions(+), 12 deletions(-) create mode 100644 crates/audio/src/vmd.rs create mode 100644 crates/audio/tests/vmd.rs create mode 100644 crates/media-testkit/src/vmd.rs create mode 100644 crates/media/src/vmd/mod.rs create mode 100644 crates/media/src/vmd/tests.rs create mode 100644 crates/media/tests/oracle_vmd.rs create mode 100644 crates/video/src/vmd/mod.rs create mode 100644 crates/video/tests/vmd/main.rs diff --git a/README.md b/README.md index c6e95bd4f..2f72adfe8 100644 --- a/README.md +++ b/README.md @@ -254,7 +254,7 @@ See the [release notes](CHANGELOG.md) for included functionality and known limit - [`craft-math`](crates/math/README.md): equations as one tree (the Office Math object model) with its syntaxes, read and written: OMML, presentation MathML, LaTeX math and the UnicodeMath linear format. - [`craft-media`](crates/media/README.md): bit-level I/O (`bitstream`), clean-room MP4/MOV (`isobmff`), Matroska/WebM (`matroska`) and MXF (`mxf`) demuxers and muxers, Ogg (`ogg`), AVI (`avi`) and - MPEG-TS/PS (`mpegts`) demuxers, from FilmCraft, and an ISO BMFF box reader. Containers only; codecs are in `craft-audio`, and + MPEG-TS/PS (`mpegts`) demuxers, from FilmCraft, a Sierra VMD (`vmd`) demuxer, and an ISO BMFF box reader. Containers only; codecs are in `craft-audio`, and video codecs will go in a separate crate. - [`craft-lottie`](crates/lottie/README.md): Lottie (Bodymovin) JSON: the document model of lottie-spec 1.0 (animation, assets, layers by type, shape items, masks, effects, text, markers) diff --git a/crates/audio/Cargo.toml b/crates/audio/Cargo.toml index d132f23da..55698d192 100644 --- a/crates/audio/Cargo.toml +++ b/crates/audio/Cargo.toml @@ -49,6 +49,8 @@ ac3-encode = ["craft-media/bitstream"] eac3-encode = ["craft-media/bitstream"] # MPEG-1/2 Layer III (MP3) encoder (`mp3`). mp3-encode = ["craft-media/bitstream"] +# Sierra VMD sound decoder (`vmd::Decoder`): 8-bit PCM and 16-bit DPCM. +vmd-decode = [] # Multi-resolution waveform overviews (`peaks`). peaks = [] # Encode AAC channels in parallel (with `aac-encode`). @@ -67,6 +69,8 @@ symphonia = { version = "0.6.1", optional = true, default-features = false, feat [dev-dependencies] craft-media-testkit = { workspace = true } +# The VMD demuxer, for the VMD sound decoder's tests. +craft-media = { workspace = true, features = ["vmd"] } proptest = { workspace = true } serde_json = { workspace = true } @@ -74,6 +78,10 @@ serde_json = { workspace = true } name = "audio" required-features = ["decode", "lossless-encode", "peaks"] +[[test]] +name = "vmd" +required-features = ["vmd-decode"] + [[test]] name = "codecs" required-features = ["decode", "encode"] diff --git a/crates/audio/README.md b/crates/audio/README.md index 5a5df4f2f..59256822a 100644 --- a/crates/audio/README.md +++ b/crates/audio/README.md @@ -42,6 +42,7 @@ No default features. The shared types (`AudioBuffer`, `AudioInfo`, `ChannelLayou | `mpa-decode` | `mpa::Decoder` | craft-media `bitstream` | | `mp3-decode` | `mp3::Decoder`, `mp3::InfoTag` | `mpa-decode` | | `mp3-encode` | `mp3::Encoder`; `encode`, `FileEncoder` for MP3 | craft-media `bitstream` | +| `vmd-decode` | `vmd::Decoder`: Sierra VMD sound (8-bit PCM and 16-bit DPCM), frame by frame from `craft_media::vmd` | none | | `peaks` | `peaks::Peaks` | none | | `rayon` | parallel AAC encoding (with `aac-encode`) | `rayon` | | `serde` | `Serialize`/`Deserialize` on the shared types | `serde` | diff --git a/crates/audio/src/lib.rs b/crates/audio/src/lib.rs index 1c6c73dcf..cd250bd21 100644 --- a/crates/audio/src/lib.rs +++ b/crates/audio/src/lib.rs @@ -27,7 +27,8 @@ //! - Codec modules for packet-level use, each with its own features: [`aac`] (`aac-decode`, //! `aac-encode`), [`alac`] (`alac-decode`, `alac-encode`), [`opus`] (`opus-decode`, and //! [`opus::encoder`] with `opus-encode`), [`ac3`] (`ac3-decode`), [`mpa`] (`mpa-decode`), [`mp3`] -//! (`mp3-decode`, `mp3-encode`), [`flac`] (`flac-encode`: the packet encoder). +//! (`mp3-decode`, `mp3-encode`), [`flac`] (`flac-encode`: the packet encoder), [`vmd`] +//! (`vmd-decode`: Sierra VMD sound). //! //! The shared types below build with no features. @@ -53,6 +54,8 @@ pub mod peaks; mod read; #[cfg(any(feature = "aac-decode", feature = "alac-decode", feature = "opus-decode", feature = "ac3-decode", feature = "mpa-decode"))] mod track; +#[cfg(feature = "vmd-decode")] +pub mod vmd; #[cfg(any(feature = "decode-native", feature = "pcm-encode"))] mod wav; #[cfg(any( @@ -224,6 +227,16 @@ impl From for AudioError { } } +#[cfg(feature = "vmd-decode")] +impl From for AudioError { + fn from(e: vmd::Error) -> Self { + match e { + vmd::Error::Unsupported(_) => AudioError::Unsupported(e.to_string()), + _ => AudioError::Malformed(e.to_string()), + } + } +} + #[cfg(feature = "opus-encode")] impl From for AudioError { fn from(e: opus::encoder::Error) -> Self { diff --git a/crates/audio/src/vmd.rs b/crates/audio/src/vmd.rs new file mode 100644 index 000000000..6c2a0b829 --- /dev/null +++ b/crates/audio/src/vmd.rs @@ -0,0 +1,178 @@ +//! Sierra VMD sound decoder (feature `vmd-decode`): the 8-bit PCM and 16-bit DPCM sound of VMD +//! (Video and Music Data) files, as in Sierra's and Coktel Vision's 1990s games. +//! +//! Clean-room, from the public format description at +//! ("Description of the Sierra Video and Music Data (VMD) Format", Mike Melanson and Vladimir +//! Gneushev). The description gives the coding but not the size of a block; the sizes here are +//! what ffmpeg was seen to expect of files made by this crate's tests: +//! +//! | Sound | Bytes in a block of `n` samples a channel | +//! |---|---| +//! | 8-bit | `n` for each channel, interleaved | +//! | 16-bit | 2 for each channel's first sample, then one code for each further sample, interleaved | +//! +//! The header's frame length is `n`, except in old-style stereo (audio flag 0x8000), where it +//! counts both channels. Old-style 16-bit stereo blocks begin with their first samples like the +//! others, which is what ffmpeg reads; the description says they carry the last block's samples +//! over instead. No such file was at hand to settle it. +//! +//! A [`Decoder`] is made from the 816-byte file header. [`Decoder::decode`] takes a sound +//! frame's flags (from its record in the file's table of contents; `craft_media::vmd` reads it) +//! and its data, and returns interleaved `i16` samples: one block of sound or silence, or, for a +//! frame with a silence mask, several. 8-bit samples are widened to 16 bits. Blocks do not depend +//! on each other, so decoding can start at any frame. +//! +//! Not decoded: IMA ADPCM sound (audio flag 0x10) and the external codecs of later Coktel Vision +//! files ([`Error::Unsupported`]). +//! +//! Input is treated as hostile: a frame yields at most 32 blocks (the width of the silence +//! mask) of at most 65 535 samples a channel, and a block is decoded only if all of it is there. No `unsafe`, no panics. + +#![warn(missing_docs)] + +/// Bytes of the VMD file header a [`Decoder`] is made from. +pub const HEADER_LEN: usize = 816; + +/// Errors from decoding VMD sound. +#[non_exhaustive] +#[derive(Debug, Clone, PartialEq, Eq, thiserror::Error)] +pub enum Error { + /// The header or a frame is not valid. + #[error("invalid VMD sound: {0}")] + Invalid(&'static str), + /// Valid, but not something this decoder handles. + #[error("unsupported VMD sound: {0}")] + Unsupported(&'static str), +} + +/// The steps of the 16-bit coding: a code's low seven bits pick one, its top bit the sign. +const STEPS: [u16; 128] = [ + 0, 8, 16, 32, 48, 64, 80, 96, 112, 128, 144, 160, 176, 192, 208, 224, 240, 256, 272, 288, 304, 320, 336, 352, 368, 384, 400, 416, 432, 448, 464, + 480, 496, 512, 520, 528, 536, 544, 552, 560, 568, 576, 584, 592, 600, 608, 616, 624, 632, 640, 648, 656, 664, 672, 680, 688, 696, 704, 712, 720, + 728, 736, 744, 752, 760, 768, 776, 784, 792, 800, 808, 816, 824, 832, 840, 848, 856, 864, 872, 880, 888, 896, 904, 912, 920, 928, 936, 944, 952, + 960, 968, 976, 984, 992, 1000, 1008, 1016, 1024, 1088, 1152, 1216, 1280, 1344, 1408, 1472, 1536, 1600, 1664, 1728, 1792, 1856, 1920, 1984, 2048, + 2304, 2560, 2816, 3072, 3328, 3584, 3840, 4096, 5120, 6144, 7168, 8192, 12288, 16384, +]; + +/// A VMD sound decoder. +#[derive(Clone, Debug)] +pub struct Decoder { + sample_rate: u32, + channels: usize, + sixteen_bit: bool, + /// Samples of one block, all channels counted. + block_samples: usize, +} + +impl Decoder { + /// A decoder for the file whose header is `header` (its first 816 bytes; more are ignored). + pub fn new(header: &[u8]) -> Result { + let h = header.get(..HEADER_LEN).ok_or(Error::Invalid("the header is shorter than 816 bytes"))?; + let u16_at = |at: usize| u16::from_le_bytes([h[at], h[at + 1]]); + if usize::from(u16_at(0)) != HEADER_LEN - 2 { + return Err(Error::Invalid("not a VMD header")); + } + if u16_at(16) & 0x1000 == 0 { + return Err(Error::Invalid("the file has no sound")); + } + let flags = u16_at(810); + if flags & 0x10 != 0 { + return Err(Error::Unsupported("IMA ADPCM sound")); + } + let length = u16_at(806) as i16; + let frame_length = usize::from(length.unsigned_abs()); + // old-style stereo counts both channels, new-style one + let (channels, block_samples) = match flags { + f if f & 0x8000 != 0 => (2, frame_length & !1), + f if f & 0x200 != 0 => (2, frame_length * 2), + _ => (1, frame_length), + }; + Ok(Decoder { sample_rate: u32::from(u16_at(804)), channels, sixteen_bit: length < 0, block_samples }) + } + + /// Samples a second. + pub fn sample_rate(&self) -> u32 { + self.sample_rate + } + + /// 1 or 2. + pub fn channels(&self) -> usize { + self.channels + } + + /// Bits a sample in the file: 8 (PCM) or 16 (DPCM). The decoder returns `i16` for both. + pub fn bits(&self) -> u8 { + if self.sixteen_bit { 16 } else { 8 } + } + + /// Samples of one block for each channel. + pub fn block_samples(&self) -> usize { + self.block_samples / self.channels + } + + /// Bytes of one block of sound in a frame. + pub fn block_bytes(&self) -> usize { + if self.sixteen_bit { (self.block_samples + self.channels).min(self.block_samples * 2) } else { self.block_samples } + } + + /// Decode one sound frame: `flags` is the frame's flag byte from its record, `data` the + /// frame data. Returns interleaved samples, [`Decoder::block_samples`] for each channel and + /// block. + /// + /// - flags 1: one block of sound; + /// - flags 2: a 32-bit mask, then blocks of sound: the sound a player loads ahead at the + /// start of a file. From bit 0 up, a set bit is a block of silence and a clear bit the next + /// block of data, while there is one. ffmpeg plays the silent blocks first instead; + /// the order here is the description's; + /// - flags 3: one block of silence. + /// + /// An empty frame gives no samples. A block cut short is an error. + pub fn decode(&self, flags: u8, data: &[u8]) -> Result, Error> { + let mut out = Vec::new(); + match flags { + 3 => out.resize(self.block_samples, 0), + _ if data.is_empty() => {} + 1 => self.block(data, &mut out)?, + 2 => { + let (mask, mut data) = data.split_first_chunk::<4>().ok_or(Error::Invalid("the silence mask is cut short"))?; + let mask = u32::from_le_bytes(*mask); + // bit by bit: a set bit is a silent block, a clear one takes the next block of data + // while there is one. The header's buffer count is not needed: game files hold + // one block fewer than it says. + for i in 0..32 { + if mask >> i & 1 == 1 { + out.resize(out.len() + self.block_samples, 0); + } else if !data.is_empty() && data.len() >= self.block_bytes() { + self.block(data, &mut out)?; + data = &data[self.block_bytes()..]; + } + } + } + _ => return Err(Error::Unsupported("unknown sound frame flags")), + } + Ok(out) + } + + /// Decode the block at the start of `data` onto `out`. + fn block(&self, data: &[u8], out: &mut Vec) -> Result<(), Error> { + let data = data.get(..self.block_bytes()).ok_or(Error::Invalid("a block of sound is cut short"))?; + if !self.sixteen_bit { + out.extend(data.iter().map(|&b| (i16::from(b) - 0x80) << 8)); + return Ok(()); + } + // each channel's first sample as it is, then a signed step for each further sample + let mut last = [0i32; 2]; + let Some((first, codes)) = data.split_at_checked(self.channels * 2) else { return Ok(()) }; + for (ch, s) in first.as_chunks::<2>().0.iter().enumerate() { + last[ch] = i32::from(i16::from_le_bytes(*s)); + out.push(last[ch] as i16); + } + for (i, &code) in codes.iter().enumerate() { + let ch = i % self.channels; + let step = i32::from(STEPS[usize::from(code & 0x7F)]); + last[ch] = (if code & 0x80 != 0 { last[ch] - step } else { last[ch] + step }).clamp(-32768, 32767); + out.push(last[ch] as i16); + } + Ok(()) + } +} diff --git a/crates/audio/tests/vmd.rs b/crates/audio/tests/vmd.rs new file mode 100644 index 000000000..d7004a888 --- /dev/null +++ b/crates/audio/tests/vmd.rs @@ -0,0 +1,265 @@ +//! VMD sound decoder tests. ffmpeg reads VMD but cannot write it, so the files come from the +//! test kit's VMD writer. The decoder must return the samples that were packed, and the same +//! samples as ffmpeg (skipped without ffmpeg). + +use std::process::Command; + +use craft_audio::vmd::{Decoder, Error}; +use craft_media::vmd::{FrameKind, open}; +use craft_media_testkit::hostile::Rng; +use craft_media_testkit::vmd::{Audio, Frame, Spec, build, dpcm_pack}; + +/// (frame length field, audio flags, channels, 16-bit): every layout the decoder handles. +const FORMATS: [(i16, u16, usize, bool); 6] = + [(50, 0, 1, false), (50, 0x200, 2, false), (50, 0x8000, 2, false), (-50, 0, 1, true), (-50, 0x200, 2, true), (-50, 0x8000, 2, true)]; + +fn spec(frame_length: i16, flags: u16) -> Spec { + let mut spec = Spec::new(4, 4); + spec.audio = Some(Audio { sample_rate: 22050, frame_length, buffers: 5, flags }); + spec +} + +fn decoder(frame_length: i16, flags: u16) -> Decoder { + Decoder::new(&build(&spec(frame_length, flags), &[])).unwrap() +} + +/// One block of random sound: its bytes in the file and the samples it decodes to. +fn block(rng: &mut Rng, d: &Decoder) -> (Vec, Vec) { + let n = d.block_samples() * d.channels(); + if d.bits() == 8 { + let bytes: Vec = (0..n).map(|_| rng.next_u64() as u8).collect(); + let samples = bytes.iter().map(|&b| (i16::from(b) - 0x80) << 8).collect(); + return (bytes, samples); + } + // a random walk, so that the steps are of every size + let mut last = [0i32; 2]; + let samples: Vec = (0..n) + .map(|i| { + let bits = 1 + rng.below(15); + let step = rng.below(1 << bits) as i32 - (1 << (bits - 1)); + let s = &mut last[i % d.channels()]; + *s = (*s + step).clamp(-32768, 32767); + *s as i16 + }) + .collect(); + dpcm_pack(&samples, d.channels()) +} + +/// A file of sound frames (with a small picture each, as ffmpeg expects video) and the samples +/// expected of it. The first frame is a mask frame: one silent block, then four of sound. +fn movie(seed: u64, frame_length: i16, flags: u16) -> (Vec, Vec) { + let spec = spec(frame_length, flags); + let d = decoder(frame_length, flags); + let mut rng = Rng(seed | 1); + let mut picture = vec![2u8]; + picture.extend([9; 16]); + let silence = vec![0i16; d.block_samples() * d.channels()]; + let mut want = Vec::new(); + let mut blocks = Vec::new(); + for n in 0..12 { + let sound = match n { + 0 => { + let mut data = 1u32.to_le_bytes().to_vec(); + want.extend_from_slice(&silence); + for _ in 0..4 { + let (bytes, samples) = block(&mut rng, &d); + data.extend_from_slice(&bytes); + want.extend_from_slice(&samples); + } + Frame::audio(2, data) + } + _ if rng.below(4) == 0 => { + want.extend_from_slice(&silence); + Frame::audio(3, vec![]) + } + _ => { + let (bytes, samples) = block(&mut rng, &d); + want.extend_from_slice(&samples); + Frame::audio(1, bytes) + } + }; + blocks.push(vec![sound, Frame::video([0, 0, 3, 3], false, picture.clone())]); + } + (build(&spec, &blocks), want) +} + +fn decode_all(bytes: &[u8]) -> Vec { + let file = open(bytes).unwrap(); + let d = Decoder::new(file.header_bytes()).unwrap(); + let mut out = Vec::new(); + for (i, frame) in file.frames.iter().enumerate() { + if let FrameKind::Audio { flags } = frame.kind { + out.extend(d.decode(flags, &file.read_frame(bytes, i).unwrap()).unwrap()); + } + } + out +} + +#[test] +fn decodes_the_samples_that_were_packed() { + for (i, (frame_length, flags, channels, sixteen)) in FORMATS.into_iter().enumerate() { + let d = decoder(frame_length, flags); + assert_eq!((d.channels(), d.bits(), d.sample_rate()), (channels, if sixteen { 16 } else { 8 }, 22050), "format {i}"); + // old-style stereo counts both channels in the frame length + assert_eq!(d.block_samples(), if flags & 0x8000 != 0 { 25 } else { 50 }, "format {i}"); + let (bytes, want) = movie(i as u64 + 1, frame_length, flags); + assert_eq!(want.len(), 16 * d.block_samples() * channels); + assert!(decode_all(&bytes) == want, "format {i}"); + } +} + +#[test] +fn matches_ffmpeg() { + let Some(ffmpeg) = craft_media_testkit::ffmpeg_or_skip("vmd sound oracle") else { return }; + let dir = craft_media_testkit::private_dir("vmd-sound"); + for (i, (frame_length, flags, ..)) in FORMATS.into_iter().enumerate() { + let (bytes, _) = movie(i as u64 + 1, frame_length, flags); + let path = dir.join(format!("sound{i}.vmd")); + std::fs::write(&path, &bytes).unwrap(); + let out = Command::new(&ffmpeg).args(["-v", "error", "-i"]).arg(&path).args(["-vn", "-f", "s16le", "-"]).output().unwrap(); + assert!(out.status.success() && out.stderr.is_empty(), "format {i}: ffmpeg: {}", String::from_utf8_lossy(&out.stderr)); + let theirs: Vec = out.stdout.as_chunks::<2>().0.iter().map(|b| i16::from_le_bytes(*b)).collect(); + assert!(theirs == decode_all(&bytes), "format {i} differs from ffmpeg"); + } + std::fs::remove_dir_all(&dir).unwrap(); +} + +#[test] +fn sixteen_bit_steps() { + // values ffmpeg gave for this block: the first sample, then signed steps from the table + let mut data = 1000i16.to_le_bytes().to_vec(); + data.extend([1, 1, 1, 0x81, 0x81, 2, 3, 0x7F, 0xFF, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1]); + let d = decoder(-20, 0); + assert_eq!(d.block_bytes(), 21); + assert_eq!( + d.decode(1, &data).unwrap(), + [1000, 1008, 1016, 1024, 1016, 1008, 1024, 1056, 17440, 1056, 1056, 1056, 1056, 1056, 1056, 1056, 1056, 1056, 1056, 1064] + ); + // stereo: the channels take turns + let mut data = [1000i16.to_le_bytes(), (-1000i16).to_le_bytes()].concat(); + data.extend([1, 0x82].repeat(3)); + let d = decoder(-4, 0x200); + assert_eq!(d.decode(1, &data).unwrap(), [1000, -1000, 1008, -1016, 1016, -1032, 1024, -1048]); + // steps past 16 bits stop at the ends + let mut data = 30000i16.to_le_bytes().to_vec(); + data.extend([0x7F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF]); + assert_eq!(decoder(-7, 0).decode(1, &data).unwrap(), [30000, 32767, 16383, -1, -16385, -32768, -32768]); +} + +#[test] +fn eight_bit_and_silence() { + let d = decoder(4, 0); + assert_eq!(d.decode(1, &[0x80, 0x00, 0xFF, 0x70]).unwrap(), [0, -32768, 0x7F00, -0x1000]); + assert_eq!(d.decode(3, &[]).unwrap(), [0; 4]); + assert_eq!(d.decode(1, &[]).unwrap(), [0i16; 0]); + assert_eq!(decoder(4, 0x200).decode(3, &[]).unwrap(), [0; 8]); + assert_eq!(decoder(-4, 0x8000).decode(3, &[]).unwrap(), [0; 4]); +} + +#[test] +fn silence_mask_runs_bit_by_bit() { + let d = decoder(2, 0); + let frame = |mask: u32, data: &[u8]| d.decode(2, &[&mask.to_le_bytes()[..], data].concat()); + let (a, b) = ([0x90, 0x91], [0xA0, 0xA1]); + let (sa, sb) = ([0x1000, 0x1100], [0x2000, 0x2100]); + assert_eq!(frame(0, &[a, b].concat()).unwrap(), [sa, sb].concat()); + // silent, sound, silent, sound + assert_eq!(frame(0b0101, &[a, b].concat()).unwrap(), [[0, 0], sa, [0, 0], sb].concat()); + // game files set a bit for every block when all are silent + assert_eq!(frame(0b0111, &[]).unwrap(), [0; 6]); + // a trailing part of a block is left out + assert_eq!(frame(0, &[0x90, 0x91, 0xA0]).unwrap(), sa); + assert_eq!(d.decode(2, &[0, 0, 0]), Err(Error::Invalid("the silence mask is cut short"))); +} + +#[test] +fn bad_input_is_an_error() { + let d = decoder(-20, 0); + assert_eq!(d.decode(1, &[0; 20]), Err(Error::Invalid("a block of sound is cut short"))); + assert_eq!(d.decode(4, &[0; 21]), Err(Error::Unsupported("unknown sound frame flags"))); + + let good = build(&spec(20, 0), &[]); + let with = |patch: &dyn Fn(&mut Vec)| { + let mut h = good.clone(); + patch(&mut h); + Decoder::new(&h).err() + }; + assert_eq!(with(&|_| {}), None); + assert_eq!(with(&|h| h.truncate(815)), Some(Error::Invalid("the header is shorter than 816 bytes"))); + assert_eq!(with(&|h| h[0] = 0), Some(Error::Invalid("not a VMD header"))); + assert_eq!(with(&|h| h[17] = 0), Some(Error::Invalid("the file has no sound"))); + assert_eq!(with(&|h| h[810] = 0x10), Some(Error::Unsupported("IMA ADPCM sound"))); + // a frame length of nothing decodes to nothing + let mut empty = good.clone(); + empty[806..808].fill(0); + let d = Decoder::new(&empty).unwrap(); + assert_eq!(d.decode(1, &[1, 2, 3]).unwrap(), [0i16; 0]); + assert_eq!(d.decode(2, &[0, 0, 0, 0, 1, 2]).unwrap(), [0i16; 0]); +} + +#[test] +fn hostile_input_never_panics() { + for (i, (frame_length, flags, ..)) in FORMATS.into_iter().enumerate() { + let (bytes, _) = movie(i as u64 + 1, frame_length, flags); + craft_media_testkit::hostile::hammer(&format!("vmd sound format {i}"), &bytes, 10_000, |d| { + let Ok(file) = open(d) else { return }; + let Ok(decoder) = Decoder::new(file.header_bytes()) else { return }; + for (i, frame) in file.frames.iter().enumerate() { + let FrameKind::Audio { flags } = frame.kind else { continue }; + if let Ok(data) = file.read_frame(d, i) { + let _ = decoder.decode(flags, &data); + } + } + }); + } +} + +/// Real files, for whoever has some: the sound of every `.vmd` in the directory +/// `CRAFT_MEDIA_VMD_SAMPLES` names must decode to the samples ffmpeg gets. Game files are not +/// redistributable, so none are in the repository and the test is skipped without the variable. +#[test] +fn real_files_match_ffmpeg() { + let Some(dir) = std::env::var_os("CRAFT_MEDIA_VMD_SAMPLES").filter(|d| !d.is_empty()) else { + craft_media_testkit::skip_test("vmd samples", "CRAFT_MEDIA_VMD_SAMPLES is not set"); + return; + }; + let Some(ffmpeg) = craft_media_testkit::ffmpeg_or_skip("vmd sound samples") else { return }; + let mut paths: Vec<_> = std::fs::read_dir(&dir) + .unwrap() + .filter_map(|e| Some(e.ok()?.path())) + .filter(|p| p.extension().is_some_and(|e| e.eq_ignore_ascii_case("vmd"))) + .collect(); + paths.sort(); + assert!(!paths.is_empty(), "no .vmd files in CRAFT_MEDIA_VMD_SAMPLES"); + let (mut files, mut samples) = (0, 0); + for path in &paths { + let name = path.display(); + let source = std::fs::File::open(path).unwrap(); + let file = open(&source).unwrap(); + if file.audio_params().is_none() { + continue; + } + let d = Decoder::new(file.header_bytes()).unwrap(); + let mut ours = Vec::new(); + for (i, frame) in file.frames.iter().enumerate() { + let FrameKind::Audio { flags } = frame.kind else { continue }; + let mut data = file.read_frame(&source, i).unwrap(); + if flags == 2 && data.len() >= 4 { + // ffmpeg plays a mask frame's silent blocks before its sound; put ours in that + // order to compare + let silent = u32::from_le_bytes([data[0], data[1], data[2], data[3]]).count_ones() as usize; + ours.resize(ours.len() + silent * d.block_samples() * d.channels(), 0); + data[..4].fill(0); + } + ours.extend(d.decode(flags, &data).unwrap_or_else(|e| panic!("{name}: frame {i}: {e}"))); + } + let out = Command::new(&ffmpeg).args(["-v", "quiet", "-i"]).arg(path).args(["-vn", "-f", "s16le", "-"]).output().unwrap(); + let theirs: Vec = out.stdout.as_chunks::<2>().0.iter().map(|b| i16::from_le_bytes(*b)).collect(); + if let Some(at) = (0..ours.len().max(theirs.len())).find(|&i| ours.get(i) != theirs.get(i)) { + panic!("{name}: sample {at} differs from ffmpeg ({} samples here, {} there)", ours.len(), theirs.len()); + } + files += 1; + samples += ours.len(); + } + println!("{files} files, {samples} samples match ffmpeg"); +} diff --git a/crates/media-testkit/src/lib.rs b/crates/media-testkit/src/lib.rs index 6d3a75525..8894014dd 100644 --- a/crates/media-testkit/src/lib.rs +++ b/crates/media-testkit/src/lib.rs @@ -13,11 +13,13 @@ //! //! [`hostile::hammer`] runs a parser over truncations and mutations of a valid input and fails on //! a panic or a slow run; `CRAFT_MEDIA_HOSTILE_MULT=10` runs ten times the iterations. +//! [`vmd`] writes Sierra VMD fixtures, which ffmpeg can read but not write. //! //! Failing loudly is what a test helper is for: a wrong `CRAFT_MEDIA_*` setting or an unwritable //! fixture directory panics, which fails the test that asked. pub mod hostile; +pub mod vmd; use std::path::{Path, PathBuf}; use std::process::Command; diff --git a/crates/media-testkit/src/vmd.rs b/crates/media-testkit/src/vmd.rs new file mode 100644 index 000000000..c7068f19c --- /dev/null +++ b/crates/media-testkit/src/vmd.rs @@ -0,0 +1,343 @@ +//! Sierra VMD fixture writer. ffmpeg reads VMD but cannot write it, so the VMD tests build their +//! files here and use ffmpeg only to decode them. Written from the public format description +//! (); the packers favour coverage of every coding +//! mode over compression. + +use crate::hostile::Rng; + +/// Bytes of the file header. +pub const HEADER_LEN: usize = 816; + +/// The audio fields of the header. +#[derive(Clone, Copy, Debug)] +pub struct Audio { + /// Samples per second. + pub sample_rate: u16, + /// The frame length field as stored: negative for 16-bit sound. + pub frame_length: i16, + /// Number of sound buffers. + pub buffers: u16, + /// Audio flags (0x8000 old stereo, 0x200 new stereo). + pub flags: u16, +} + +/// What the header says. +#[derive(Clone, Debug)] +pub struct Spec { + /// Picture width in pixels. + pub width: u16, + /// Picture height in pixels. + pub height: u16, + /// Left coordinate of the picture. + pub left: u16, + /// Top coordinate of the picture. + pub top: u16, + /// 256 R-G-B triplets of 6-bit components. + pub palette: [u8; 768], + /// Size of the unpack buffer for LZ-packed frames. + pub unpack: u32, + /// Video codec version (header bytes 4-5). + pub codec_version: u16, + /// The sound, if the file has any. + pub audio: Option